A novel UAV-integrated deep network detection and relative position estimation approach for weeds

This paper aims at presenting a novel monocular vision–based approach for drones to detect multiple type of weeds and estimate their positions autonomously precision agriculture applications. The methodology is based on classifying detecting the using proposed deep neural network architecture, named fused-YOLO images acquired from camera mounted unmanned aerial vehicle (UAV) following predefined elliptical trajectory. detection/classification complemented by new estimation scheme adopting unscented Kalman filter (UKF) exact location weeds. Bounding boxes are assigned detected targets (weeds) such that centre pixels bounding box will represent target. extracted converted into world coordinates forming azimuth elevation angles target UAV, used extract Experiments were conducted both indoor outdoor validate this integrated detection/classification/estimation approach. errors in terms misclassification mispositioning minimum, convergence position results was short taking account affordable platform with cheap sensors experiments.